Inversion of heavy metal content in soil using hyperspectral characteristic bands-based machine learning method

Area of Science:

• Environmental Science
• Geoscience
• Data Science

Background:

• Heavy metal pollution in soil poses significant environmental risks.
• Accurate monitoring of soil heavy metal content is essential for environmental protection and risk assessment.

Purpose of the Study:

• To develop and evaluate an advanced inversion model for predicting six soil heavy metals (As, Cd, Cr, Cu, Ni, Pb).
• To investigate the efficacy of hyperspectral data combined with machine learning, particularly ensemble learning, for soil heavy metal analysis.

Main Methods:

• Utilized hyperspectral data from 21 soil reference materials across China.
• Employed preprocessing algorithms and Random Forest (RF) for feature band selection.
• Developed an integrated learning (Stacked RF) model with XGBoost, LightGBM, CatBoost as base learners and RF as the meta-model.

Main Results:

• The Stacked RF model demonstrated superior accuracy and stability compared to traditional machine learning methods.
• Ensemble learning models, specifically the MF-RF-Stacking model, achieved the best performance in heavy metal inversion.
• Identified key hyperspectral characteristic bands crucial for accurate soil heavy metal prediction.

Conclusions:

• Machine learning, especially ensemble methods, offers a powerful approach for soil heavy metal inversion.
• The proposed MF-RF-Stacking model provides a robust and accurate solution for monitoring soil heavy metal contamination.
• This research offers a new perspective on utilizing hyperspectral data and ensemble learning for environmental soil analysis.